Statistical Methods for Improving Real-Time Public Health Surveillance and Integrated Outbreak Detection

点击翻译按钮获取中文摘要

Project Summary/Abstract The COVID-19 pandemic has accentuated the need for strong monitoring and surveillance systems. To conduct early detection and response to emerging infectious diseases, there must be robust analytical tools that examine historical and current data in order to identify potential aberrations in key health indicators. This is especially needed when reliable testing and reporting data is lacking. Instead, key associated indicators, namely mortality and related symptoms to a disease of interest, can be tracked and analyzed. Two problems exist: (1) reporting delays lead to undercounts in current health indicators data, and (2) prior anomalies such as spikes in mortality due to past outbreaks distort historical or baseline data. Thus, the goal is to develop methods to conduct ongoing, rolling surveillance and outbreak detection in the context of these two issues. Two large datasets resulting from collaborations are available: (1) state-level mortality data from the Centers for Disease Control and Prevention (CDC) and Departments of Public Health (DPH) in Puerto Rico, Massachusetts, and California from January 2017-December 2021, and (2) Partners in Health (PIH) routinely collected health management information systems (HMIS) data on COVID-19-associated indicators, specifically acute respiratory infections (ARI) from 900 health facilities in 8 countries from January 2016-current. Through the first aim of the proposed research plan, the first dataset will be analyzed to develop methods for imputing undercounts in current data. In doing so, various methodological gaps in existing research will be addressed, including accounting for seasonality in reporting lag patterns and providing measures of uncertainty around estimates. Through the second aim of the proposed research plan, the second dataset set, along with a simulated version, will be analyzed to develop methods for rolling outbreak detection by simultaneously addressing two gaps: accounting for prior data aberrations and optimizing key statistical properties including bias, variance, and appropriate model fit. Both goals are complementary and equally important in infectious disease surveillance. While the specific datasets and indicators as described above will be analyzed, the developed methods will be broadly applicable to monitoring of any key health indicators. As COVID-19-related challenges persist and new threats emerge, statistically rigorous tools for early detection remain of paramount importance. Just as important is dissemination of these tools in accessible, easily usable open-source software packages, a key aspect of the proposed research plan.